Landslide susceptibility mapping at Ovacık-Karabük (Turkey) using different artificial neural network models: comparison of training algorithms

摘要: This study aims to investigate the performances of different training algorithms used for an artificial neural network (ANN) method produce landslide susceptibility maps. For this purpose, Ovacik region (southeast Karabuk Province), located in Western Black Sea Region (Turkey), was selected as area. A total 196 landslides were mapped, and a database prepared. Topographical elevation, slope angle, aspect, wetness index, lithology, vegetation index parameters taken into account analyses. Two ANN structures, which composed single double hidden layers, applied compare effects ANN. Four algorithms, namely batch back-propagation, quick propagation, conjugate gradient descent (CGD), Levenberg–Marquardt, stage models. Thus, eight maps produced area using structures algorithms. In order assess spatial considered on models, relative operating characteristics (ROC) relation value (rij) approaches used. The map by CGD1 has highest AUC (0.817) rij values (0.972). Comparison indicated that CGD algorithm is slowest one among other but showed performance results.